Colloquium: Distributed Bayesian Inference for Big Survival Data

Dr. Ray Bai will be presenting at 11 AM in the Z. Smith Reynolds (ZSR) Auditorium, Room 404. Dr. Ray Bai is an Assistant Professor of Statistics at the University of South Carolina. He earned his PhD in Statistics from the University of Florida in 2018 under the supervision of Dr. Malay Ghosh. He then completed a two-year postdoc at the University of Pennsylvania before joining University of South Carolina as faculty in 2020. His research interests include Bayesian statistics, deep learning, and scalable algorithms for high-dimensional and complex data.

Dr. Bai’s talk is entitled “Distributed Bayesian Inference for Big Survival Data.” An abstract is as follows. 

Abstract: Bayesian methods for survival analysis are attractive because of their flexibility and their ability to provide natural inference for the covariate effects and other nonparametric quantities of interest (e.g. the survival function) through the posterior distribution. However, fitting Bayesian methods with Markov chain Monte Carlo (MCMC) is challenging – and possibly infeasible – when the sample size N is large. To alleviate this computational burden,  we propose distributed Bayesian inference algorithms for survival analysis based on piecewise exponential (PWE) models. We theoretically justify our approach by showing that under some conditions, posterior inference under our distributed method is asymptotically equivalent to posterior inference under the true posterior. Next, we extend our algorithm to the mixed effects PWE model for clustered survival data. For clustered data, we propose a novel data partitioning scheme for the division step. Our methods are illustrated on a kidney transplantation dataset from the Organ Procurement and Transplantation Network (OPTN)  with N=194,246 patients from 287 kidney transplantation centers.